RealSSA Benchmark for Streaming Audio-Visual Target Speaker Extraction
October 22, 2026
The RealSSA benchmark introduces simulated and real-world acoustic scenes featuring 1–5 dynamic near-field interfering speakers and up to 15 far-field sources. It addresses the lack of realistic, multi-speaker, and reverberant test data for causal audio-visual extraction systems.
HOW THIS AFFECTS YOU
●
researcherYou can use this to evaluate causal architectures against high-interference, real-world noise profiles.